feat(engine): load corpus/context/ at startup for calendar and hook data

[ehotting]

Context

PR feat/milieurecht introduces corpus/context/ — a directory for domain knowledge that is not law but needed to execute law. Currently it contains:

• nl/calendar/ — Dutch public holiday dates per year (2020-2035)
• nl/hooks/ — Hook registers for cross-cutting laws (ATW, AWB)

The BDD test harness reads these files directly. The engine does not load them yet.

What needs to happen

The engine should load corpus/context/ at startup alongside corpus/regulation/, making context data available during law execution. This enables:

1. Automatic calendar injection: When the Termijnenwet hook fires, the engine can automatically look up holiday dates from the calendar files instead of requiring the caller to pass all 9 dates as parameters.

2. Hook register validation: The engine can validate that hooks declared in law YAML files match the hook register, catching configuration errors early.

3. Context-aware execution: Future context data (institutional structure, scope dimensions) can be loaded the same way.

Design considerations

• Context files have different schemas than regulation files — need a separate ContextData struct, not ArticleBasedLaw
• Add context_base_path() alongside regulation_base_path() (already added in BDD test common)
• Calendar data is keyed by year — engine needs to resolve the correct year from the execution date
• Hook registers are informational — engine may validate against them but they don't drive execution (the hooks in the law YAML do)

References

• RFC-015: Engine Policy (describes the three-way split: regulation / context / engine-policy)
• corpus/context/README.md for full explanation